AI agents read your launch page before humans do

A growing share of the first visits to a newly launched tool are not people. They are assistants fetching your page to answer somebody's question about which tool to use, or coding agents checking whether your API exists before recommending it. They arrive, fetch once, and leave with whatever the raw HTML contained.

This is not a future scenario. Anthropic's Claude Cowork moved to web and mobile on 2026-07-07 according to TechCrunch's report, and agent-shaped products are now a regular category on launch boards, including this one.

• Agent tools by launch date

What breaks when a crawler does not run JavaScript

Most major crawlers used by AI systems fetch HTML and do not execute client-side code. So a pricing table assembled in the browser, a feature list injected after load, or a FAQ built by a component that runs on mount are all invisible to them, even though your page looks complete to you.

The fix is unglamorous: render the facts server-side. Prices, limits, supported formats and the one-line description of what the product does should be present in the document that arrives before any script runs. You can check with a single fetch and a search for the sentence you expect to find.

• The sentence that needs to be in the HTML

What Google says you do not need to build

Google's AI features guidance, updated 2026-07-10, states that you do not need new machine readable files, AI text files, markup or Markdown to appear in Search, and that such files will neither harm nor help visibility or rankings. It also says structured data is not required for generative AI search and there is no requirement to split content into small pieces.

That is a useful budget saver during a launch week. The files are cheap to publish and fine to keep for other consumers, but they are not the lever, and anyone selling them to you as a ranking tactic during your launch is selling something Google has explicitly disclaimed.

What does help

None of that is separate from writing a good page for people. It is the same page, with the facts placed where a fetch can see them.

• Facts in server-rendered HTML at a stable URL, including the numbers in your free tier.
• A page that names what the product cannot do, since that is what an assistant needs to answer a comparison question honestly.
• Documentation reachable without a login, because an agent that cannot read your API reference will recommend one it could read.
• A listing on an indexable directory with a real launch date, which is one more retrievable page describing your product in a form a machine can quote.

• Publish a retrievable listing
• Making the listing worth quoting

The launch-week version of this

In the week before you announce, run one check: fetch your own pages the way a crawler would, without a browser, and confirm the price, the limits and the one-liner are all in the response. Makers who do this routinely find that the single most quoted fact about their product, the price, was assembled client-side and therefore did not exist to anyone reading the page programmatically.

• Where the check fits in the pre-flight list
• What gets quoted once the page is readable